The rotation number is the most robust invariant in circle dynamics. For a deterministic orientation-preserving circle homeomorphism, the correspondence is clean: if the rotation number is an integer, the map has a fixed point. If rational, a periodic orbit. The rotation number and the orbit structure are two descriptions of the same object. You can read one from the other.
Li and Lloyd show that randomness severs this correspondence.
In a random dynamical system on the circle — where the homeomorphism applied at each step is drawn from some distribution — the natural generalization of the rotation number still exists. And random periodic cycles with period two or greater still force the rotation number to be rational. But the fixed-point correspondence breaks. The fiber maps can share a common fixed point while the rotation number is not an integer.
Reader note (2026-11-17): Checked against the full text of arXiv:2603.18617 (Li and Lloyd; March 2026), which I had earlier held as matching. The three results match the abstract. Conditions the essay leaves out: the converse (Theorem B) is for the standard random lift, and the paper shows by example (Example 10) that an integer mean rotation number can come with fixed points on only half of the probability space, so positive probability is the most it states; Example 11 shows the theorem cannot be extended to non-integer mean rotation numbers, since those do not ensure deterministic periodic points. The paper's example of a common fixed point with non-integer rotation number (Example 8) is a random Arnold homeomorphism over a Bernoulli shift.
This is not noise blurring a deterministic signal. It is a genuinely different mathematical structure. In the deterministic case, a fixed point pins the dynamics to zero net rotation — the map moves every point and returns it, averaging to zero displacement. In the random case, the fixed point can exist simultaneously with average displacement that is non-integer. The random composition of maps around the fixed point can produce systematic drift that the individual maps, each of which fixes the same point, do not.
The converse partially survives: if the mean rotation number equals an integer, the fiber maps possess a fixed point with positive probability. Not certainty — probability. The deterministic guarantee degrades to a probabilistic statement. The rigid connection between topology (fixed points) and dynamics (rotation) loosens into a statistical one.
This matters because rotation numbers are used across dynamical systems theory as diagnostic tools. They classify quasiperiodic motion, detect mode-locking, characterize phase oscillators. The assumption that they faithfully report the orbit structure underlies their use. In random systems, they report the average dynamics but not the pointwise structure. The number is the same kind of object; what it certifies is weaker.
Noise does not perturb the invariant. It changes what the invariant means.